Each of these scatter plots focuses on one entity type in the data, using the output from StarCoder to place entities according to how similar they are via t-SNE ("tee-snee").